Fig. 4

a Schematic description of methods 3 and 4: the anchor is extended one residue before and after, and for each residue, backbone torsions are sampled. For each backbone geometry, the interface metrics are calculated for different amino acid substitutions for that residue. Inset: Example of ∆∆G distribution for a single residue position for different backbone geometries and amino acid choices. Backbone phi/psi distribution were sampled on 30˚ grids (each row is a different phi/psi bin), and the free energy of HDAC binding computed for different amino acid possibilities (y axis). ∆∆Gs are indicated in colors from light yellow (most favorable) to dark blue (most unfavorable). The best combinations of torsion and amino acid are then used for extension of the peptide sequence, closure, and design. b Chemical structure of des3.3.0 and c its computational model at the HDAC2 interface. d des3.3.0 has an IC₅₀ of 9.1 nM for HDAC2 and 12 nM for HDAC6 (Source Data are provided as a Source Data file). e Chemical structure of des4.3.1. f des4.3.1 inhibits HDAC6 with an IC₅₀ value of 17 nM, 88 times better than its potency for HDAC2 (Source Data are provided as a Source Data file). g Crystal structure of bound des4.3.1 (PDB ID: 6WSJ, light gray) is different from the designed model (dark gray). d-Arg8 (shown as sticks) adopts a negative phi torsion, a geometry more consistent with l-Arg. h Crystal structure of des4.3.1 (PDB ID: 6WSJ) complexed with HDAC6. The HDAC6 structure shows minimal change upon binding (RMSD of 0.17 Å for 302 Cα atoms compared with the apo-structure, PDB ID: 5EEM). The non-interacting residues are shown as lines and their side-chains are omitted for clarity. Water molecules are shown as green spheres.